Income, Inequality and Food Prices

8/8/2019 Income, Inequality and Food Prices

1/29

Income,Inequality,andFood

Prices:ACritiqueofBroda,Leibtag,andWeinsteinsTheRoleofPricesin

MeasuringthePoorsLivingStandards

ShawnFremstad

0BDecember2010

CenterforEconomicandPolicyResearch

1611ConnecticutAvenue,NW,Suite400

Washington,D.C.20009

2022935380

www.cepr.net


2/29

CEPR Income, Inequality, and Food Prices i

Contents

UIntroduction and SummaryU ..............................................................................................................................1

U

Summary of Broda et al.U

....................................................................................................................................3 UThe Nielsen DatasetU......................................................................................................................................3

UBroda et al.s AnalysesU ...................................................................................................................................3

UIs the Nielsen Dataset Representative of Low Income Shoppers?

U............................................................7

UEven If the Nielsen Dataset is Representative of Low-Income Shoppers, Can We Conclude that thePoor Pay Less for Food? U ............................................................................................................................13

UIncome Differences in Food Prices and InequalityU ....................................................................................14

UThe Question of Whether the Poor Pay More Today: Food, Transportation and Financial Services U16

UFood U ..............................................................................................................................................................16

UTransportation

U.............................................................................................................................................17

UFinancial ServicesU ........................................................................................................................................18

UAre the Current Federal Poverty Thresholds (Way) Too High? U...............................................................19

UConclusion U ........................................................................................................................................................23

UReferencesU .........................................................................................................................................................24

About the Author

Shawn Fremstad is Director of the Inclusive and Sustainable Economy Initiative at the Center forEconomic and Policy Research in Washington, D.C.

Acknowledgments

The author thanks Dean Baker, Zo Neuberger, John Schmitt, and Nicole Woo for their helpfulcomments, Kris Warner for editing and formatting this document, Julie Strawn for her help withliteracy data, Chen Zhen for providing Nielsen demographic data, and the Annie E. CaseyFoundation for their financial support. Responsibility for any errors rests solely on the author.


3/29

CEPR Income, Inequality, and Food Prices 1

IntroductionandSummary

In The Role of Prices in Measuring the Poors Living Standards, Christian Broda, EphriamLeibtag, and David E. Weinstein (2009) use proprietary datathe 2005 Nielsen Homescandatasetto analyze differences by income level in the prices paid for food. They find that Nielsenhouseholds with incomes above $60,000 pay somewhat more for the same food items than mosthouseholds with lower incomes, with Nielsen households with incomes above $100,000 paying themost. Based on this finding and additional regression analyses, they conclude broadly that the poorpay lessnot morefor the goods they purchase and that not accounting for this suggests thatincome inequality may be between 2.5 to 5 percent less than shown by national statistics.

This paper reviews the methodology and findings of Broda et al. and concludes that:

1) Research that relies on Nielsen data to provide information about the relative prices paid bylow-income households in the population, including Broda et al.s conclusion that the poorpay less, is unlikely to be reliable. Both low- and high-income households appear to be

underrepresented in the weighted and unweighted versions of the Nielsen dataset. Moreover,both the low-income and high-income households in the dataset are almost certainlyunrepresentative of their counterparts in the population as a whole. In fact, the methodologyalmost seems designed to select the subset of lower-income households who are carefulcomparison shoppers, and to exclude those lower-income households who are likely to havethe greatest barriers to economizing on food, particularly those with disabilities, low literacylevels, and limited or no access to the internet. These differences could account entirely forBroda et al.s main finding that lower-income households pay less than higher-income onesfor food.

2) Even if we ignore the fact that Nielsen households are almost certainly unrepresentative of

households in the population as a whole, Broda et al.s interpretation of their results asshowing that the poor pay less than higher-income households for food is overstated. Inaddition to the income-price differences noted above, their results also show thathouseholds in the two lowest-income categories in the Nielsen data set (households withincomes under $5,000 and those with incomes between $5,000 and $7,999) appear to paymodestly more for food than most other low- and moderate-income households withincomes just above them. Based on the results they present, we cant say whether these twoincome categories in the Nielsen data represent a modest share of the officially poor or amore substantial one, but clearly some significant share of the poor pay more for food intheir results. Moreover, a comparison of the income categories in the Nielsen dataset withthose from other representative national surveys suggests that the bottom-two incomecategories in the Nielsen database likely represent a substantial share of officially poorhouseholds.

3) Broda et al.s poor pay less interpretation of their results is also overstated because itdoesnt account for additional costs that are likely to be disproportionately borne byeconomizing lower-income householdsthese costs include time spent comparisonshopping, time spent traveling to more distant stores, and inconveniences associated withbuying in bulk (including storage costs and wastage)or for the additional benefits relatedto store amenities that are almost certainly disproportionately realized by higher-income


4/29


households. Thus, even if we again ignore the fact that Nielsen households are almostcertainly unrepresentative of households in the population, we should be very cautious aboutconcluding that the differences Broda et al. find in the prices paid per food item byhousehold income level correspond to differences of a similar magnitude in the welfare orliving standards associated with those items.

4) While Broda et al. conclude that higher food prices paid by higher-income households meanthat income inequality is overstated in standard income statisticsthat is, that incomedifferences in food prices should be viewed as a factor that moderates the real level ofinequalityit seems more likely that the massive increase in income inequality over the lastseveral decades has directly contributed to income-based differences in food prices. Between1979 and 2003, the real incomes of families in the top fifth of the income distribution grewby 49 percent. By comparison, the incomes of families in the bottom fifth increased by onlyone percent, and those in the middle quintile by only nine percent. As a result, low- andmoderate-income households have likely felt greater pressure to economize in food andother areas than they would have if their incomes had grown at a similar rate as incomes didfor higher-income households.

5) Whether or not the poor pay more for food purchased for consumption at home is today arelatively small part of the broader and now decades-old question of whether the poor paymore. In 1960, food expenditures accounted for about one-quarter of the average familysexpenditures; today, they have dropped to nearly half that (about 13 percent). The foodexpenditure share of low-income households today (16.2 percent) is slightly higher than theaverage for all households, but the trend is the same. Researchers seeking to answer thequestion of whether the poor pay more today need to consider other factors, particularlytransportation costs and financial services, that have taken on increased importance in recentdecades.

6) Higher-income households likely benefit disproportionately from the now-widespreadacceptance of credit cards by stores selling groceries. Grocery stores neither provide adiscount for shoppers who pay cash nor require credit-card users to pay an additionalamount to cover the costs to food retailers of accepting their credit cards (roughly $2 forevery $100 of credit-card purchases). As a result, these costs are passed on in the form ofhigher prices for all food shoppers. To the extent that higher-income households are morelikely to use credit cards for grocery purchases than lower-income householdswhich seemsquite likely given general data on income differences in the use of credit cardsthe generallypoorer customers who dont use credit cards (and do not receive the benefits associated withthem) are subsidizing the purchases of the generally richer customers who do.

7) In a short section of their paper, Broda et al. argue that the official poverty rate overstatesactual poverty today by some 60 percent because of biases in the Consumer Price Index.The current poverty line implied by this findingroughly $11,025 for a family of two adultsand two children, rather than the current $22,050strains credulity. There is no reason tobelieve that a family of four can live a minimally decent life in 2010 on $11,025 a year.

The following section of this paper summarizes Broda et al.s methodology and findings. Section IIexamines the representativeness of the Nielsen dataset, especially in regard to low-incomehouseholds. Section III discusses whether Broda et als findings actually support their conclusion


5/29


that the poor pay less, even if we ignore the unrepresentativeness of the Nielsen dataset. SectionsIV and V raise some additional issues related to the relationship between inequality and food prices,the distribution of the costs and benefits of credit card use for food purchases and the extent towhich the poor pay more for financial services. Finally, Section VI discusses the distinct issue, raisedbriefly by Broda et al., of whether the current poverty thresholds ($22,050 for a family of four) are

more than double what they should be.

SummaryofBrodaetal.TheNielsenDataset

To examine the relationship between prices and household income, Broda et al. use household-leveldata from the 2005 Nielsen Homescan dataset. This dataset contains information on foodpurchases made by 40,000 U.S. households.

A strength of the Nielsen dataset is that it includes price data, household income, and some otherhousehold characteristic data, and allows for direct comparisons of the prices paid for the sameitems (food products with the same UPC codes). However, both the price and household incomedata have considerable limitations. The price of each food item purchased by a household is eitherimputed by Nielsen based on average weekly prices paid at the store or directly entered by thehousehold. As discussed further below, both processes for recording prices paid are subject to error.

The household income data in the Nielsen survey is relatively limited and imprecise: households aresimply given a list of income ranges (shown in the x axis ofFigure 1 on the next page) and asked toselect the one that their household income falls into. This method is unlikely to result in reliableincome estimates for lower-income households. Even in sophisticated public surveys that rely onprofessional interviewers using computer-assisted interviewing tools to collect data and include anarray of detailed questions about specific income sources, income tends to be underreported amonglow-income households, particularly income from public transfers. F1F

Another major weakness of the Nielsen dataset, discussed in detail in the next section of this paper,is that it is unlikely to be representative of the U.S. population both because of the selection processand the considerable burdens involved in participating. Participants in the dataset come from a self-selected pool of people who submitted applications to Nielsen to participate. Participants who areselected by Nielsen from this pool receive a hand-held device to record grocery purchases. To beincluded in the specific dataset used by Broda et al., participants must have used the device toprovide food-purchase data to Nielsen on a weekly basis for at least 10 months in 2005.

Brodaetal.sAnalyses

Broda et al. start by regressing the log of the price per food unitthe log of the price either directlyrecorded by the household or imputed by Nielsenon a series of dummy variables for the

1 On this point, see, e.g., Meyer, Mok, and Sullivan (2009) who document substantial underreporting in the 2004 CPS ofTANF income supplements (42 percent unreported), unemployment insurance (25 percent unreported), SupplementalSecurity Income (18 percent unreported), workers compensation (54 percent unreported), and the Earned IncomeTax Credit (35 percent unreported).


6/29


household income ranges in the dataset and fixed effects for each food item. The food items in thisregression appear to include both random-weight items (i.e., fresh fruits, vegetables, meat, and otheritems with no UPC) and non-random-weight items (i.e., items with a UPC barcode), with the vastmajority of items in the data being ones with UPCs.F2F

FIGURE 1

Relationship between Income and Prices Paid

Source: Broda, Leibtag, and Weinstein (2009), p. 83.

The figure above, from their paper, shows the percentage difference in average price paid per itemrelative to the price paid by households with income under $5,000 (that is, the coefficients on eachof the income-range variables in their regression). According to this analysis:

Households with incomes in the ranges between $8,000 and $45,000 appear to pay aboutone percent less per item than households with income under $5,000.

2 Although the text is somewhat confusing on this point. The label for the y-axis of Figure 1 above, which is copieddirectly from the paper, refers to price paid per UPC item. However, they subsequently note that fresh produce is theone exception to the basic rule in the data that the quality of a good is identical for all goods sold with the samebarcode.


7/29


Households with incomes above $70,000 appear to pay the most per food itemmore thantwo percent more than households with incomes in the $8,000-$45,000 income range.

Households with incomes below $8,000 appear to pay between .5 to 1.3 percent more thanhouseholds with incomes between $8,000 and $30,000.

Based on this regression, Broda et al. conclude the conventional wisdom that the poor pay higherprices is not present in a dataset that precisely tracks purchases of individual goods by differenthouseholds.

They then run an additional series of regressions to determine why this pattern of differences exists.In each of these regressions, the dependent variable is the log of the price per good, and theexplanatory variables include household income range and a fixed effect for each good. In this seriesof regressions, income range is recoded for every household by averaging the top and bottom ofeach income range. So, all households in the category of $12,000 to $15,000 are given an incomeequal to $13,500, and all households in the category of $70,000 to $99,999 are given an incomeequal to $85,000. They drop the highest-income category ($100,000 and above) from theseregressions, because the midpoint of $100,000 and up is not well-defined. These regressions alsoappear to include mostly UPC items and a small percentage of non-UPC ones.

The results for these regressions are summarized inTable 1 below. As Table 1 shows:

In a regression that includes just log household income and food item fixed effects (columnone), they find that a 10 percent increase in income is associated with roughly a 0.1 percentincrease in price per item.

They then build on this model by adding in additional controls and dependent variables(column two). Controlling for household characteristics (household size, race, marital status,age of adults) slightly increases the elasticity of price paid by household income (from 0.011to 0.013).

When the log of the per capita income of the zip code in which the household lives is addedas a dependent variable (column three), the coefficient is positive, with households that livein zip codes that are two standard deviations poorer than the average zip code paying about1.4 percent less for the same items than households that live in zip codes that are twodeviations above average. In other words, controlling for household income and otherhousehold characteristics, households living in richer neighborhoods tend to pay more forthe same food items than households living in poorer ones.

They then add the log of per capita income of the zip code in which the store is located(column four). The resulting coefficient for this variable suggests that a store in aneighborhood with twice the per capita income of another neighborhood only has pricesthat are one percent higher.

In column five, they interact household income with the log per capita income of the zipcode in which the household lives. The coefficient for the interaction term is positive,


8/29


showing that low-income people living in poor neighborhoods tend to pay less for thesame items as low-income people living in high-income neighborhoods.

In columns six and seven, they include controls for the retail chains that goods werepurchased at, which they argue is likely to eliminate the impact of unobserved quality

differences across chains on the prices paid by consumers. This addition reduces thecoefficient on the log of household income to about .0089, an association about one-thirdlower than for log household income in column two (.013), which doesnt control for retailchain. They argue that the difference between these two coefficients suggests that aboutone-third of the higher price paid by richer households for the same good is attributable tothem shopping at nicer stores while the rest is due to their shopping behavior within stores.The fact that poorer households pay less for the same goods even in the same retail chainindicates that the price differential between poor and wealthy households is likely torepresent a shopping effect. (As discussed further below, controlling only for retail chain isa fairly limited control for store quality. It seems quite likely that there are considerablequality differences between stores within retail chains.)

TABLE 1

Impact of Income on Prices Paid

Source: Broda, Leibtag, and Weinstein (2009), p. 84.

In a subsequent set of regressions, they separate items with UPCs from those without. Here, therelationship between income and prices for UPC items is similar, but somewhat smaller, than therelationship in the regressions described above. For the non-UPC (non-identical) produce items, the


9/29


coefficient for log household income is considerably higherabout five times higher than it is forUPC-only items. They conclude that the greater sensitivity to income of prices for fresh produce ismostly due to the poor finding lower-quality versions of the same produce.Finally, to further investigate the role of quality variation in household purchases, they regressed thelog average price paid by households for all barcode items in 128 product groups (these included

distinct groups such as vitamins, milk, soup, and pain remedies) on the log of the householdsaverage expenditure per adult.F

3F

The resulting coefficient shows how the average price paid forproducts in each group varies by income. They found that the median of a one log unit increase inincome is an 8.6 percent increase in the amount paid in each product group and that while the richonly pay slightly more for precisely the same type of eggs or milk or cheese ... than poorhouseholds, they spend much more for these items in general, a result that almost surely arisesfrom higher-income households purchasing higher-quality [items in each product group] thanpoorer households.

IstheNielsenDatasetRepresentativeofLowIncome

Shoppers?

As Broda et al. acknowledge in a footnotebut do not otherwise addressthere is good reason tobe concerned that the Nielsen dataset is not representative of low-income households. F4F Otherresearchers have also raised this concern. For example, according to Steward and Blisard (2008, p.155), low-income households ... appear in the [Homescan] sample in a proportion less than theirshare of the total population. However, there has been no systematic analysis of this problem.

Why might low-income households be underrepresented in the Nielsen data, and how might thatbias Broda et al.s results? The data most commonly used in price research is public datadata thatis publicly funded and produced, and publicly available at no costthat relies on random samples.By contrast, the Nielsen dataset is a proprietary one that does not utilize a random sample. Instead,as Harris (2005) explains, Nielsen uses direct mail and email to recruit participants, with the vastmajority (80 percent) of those selected invited by email.

Although Nielsen selects participants from the pool of interested applicants based on thedemographic targets (listed in the text box below), these are relatively limited (for example, there areonly four income-level targets), and wont necessarily result in a participant pool that is adequatelyrepresentative of low-income households. Once selected, participants receive financial incentivesgift points that can be redeemed for various goods and sweepstakes entriesto encourage scanning.To be included in the Nielsen data used by Broda et al., participants must have scanned their grocerypurchases on a weekly basis for at least ten months in 2005. Based on Nielsens estimate that

participation will take approximately one hour per week, complying with this requirement means

3 For this regression, they used a separate set of Nielsen data, one from 2003 that has a much smaller sample, but coversa wider range of goods than food.

4 In the text of their article, Broda et al. describe the Nielsen dataset as demographically representative, but they provideno specific information on the demographic (or geographic) representativeness of the data. In the footnote, theyacknowledge that there is some concern that Homescan data underrepresent households in the lowest part of theincome distribution and that additional research is needed to estimate how this underrepresentation, if it exists mayaffect our results.


10/29


that a household would have needed to have spent at least 43 hours in 2005 scanning and enteringfood purchases.

A truly random and representative sample of consumers will likely look considerably different froma group of consumers who: 1) learn about the Nielsen survey via email, direct mail, or word ofmouth; 2) volunteer to participate; 3) make it through the two questionnaires that are required

The Nielsen Homescan Survey Process

Nielsen uses direct mail and email to recruit Homescan participants. People interested inparticipating (even if they are not directly recruited by Nielsen) can also apply for participation on theHomescan website. To be considered for participation, potential panelists must first complete a two-

sided single sheet questionnaire on demographics. Once they return the questionnaire, they are sent afour-page booklet with additional questions. Households that return this questionnaire are put in thereserve pool.

Participants are selected from this pool using the following geographic and demographic targets:

household size (grouped into four levels)

household income (grouped into four levels)

household-head age (grouped into four levels)

female-head education (grouped into four levels)

male-head education (grouped into four levels)

presence of children (present or not present)

race (white, black, and Oriental [sic])

Hispanic (yes or no)household-head occupation (three levels)

The sample includes two panels: households that record all UPC-coded food purchases and a subsetof these households (known as the Fresh Foods panel) that record both UPC-coded purchases andnon-UPC coded purchases.

Once selected, households are provided with a handheld scanner and instructions. Participatinghouseholds scan the bar codes of UPC-coded food purchases, enter the quantities purchased, recordthe date and store of each purchase, and note whether the item was purchased at the regular orpromotional price, and whether they used a coupon for the purchase. If the purchase is made at astore that is part of Nielsens Scantrack system, Nielsen imputes the price paid using the average

weekly price paid at the store for the item. If the purchase is made at a non-Scantrack store,households enter the price directly. Nielsen checks this entered price by comparing it to a range ofprices observed elsewhere for the same or similar. If the price is out of that range, the medianregional price is substituted.

Participants in the Fresh Foods panel track UPC purchases in this same way, but also track purchasesof produce and other non-UPC random-weight food by directly recording product, weight or count,and price information.

Participants earn gift points for scanning and transmitting UPC barcodes; these points can beredeemed for electronic equipment, household items, jewelry, toys, and other items in Nielsens giftcatalog. In addition, for every week a participant scans and sends purchase information to Nielsen

they earn an entry into a sweepstakes that offers cash prices, vacations, and automobiles as prizes.


11/29


before being added to the pool; 4) once selected by Nielsen, spend some 40 hours scanning andrecording their food purchases on a weekly basis over a 10-month period; and 5) respond to thefinancial incentives offered by Nielsen for continued participation. In particular, the consumers inthe Nielsen dataset may be more likely to comparison shop and economize than a truly randomsample of consumersand the propensity to engage in comparison-shopping by income level in the

Nielsen dataset may vary from that in the overall population. So, for example, if the lower-incomeconsumer in the dataset are more likely to comparison shop than lower-income consumers in thepopulation, while the higher-income consumers in the dataset are more similar to their counterpartsin the population, research using the data will likely show a greater difference in food purchaseprices between lower- and higher-income households than actually exists.

In considering whether the Nielsen data is representative of officially poor households, it is useful toconsider some of the differences between these households and higher-income ones.

Households with incomes below the poverty line are much more likely to have one or moremembers with a disability than higher income households. For example, She and Livermore(2009) find that almost half of working-age adults living below the poverty line have a workdisability, a rate about twice that of the population overall. F5F Given the Nielsen recruitmentprocess, the lack of any demographic targets for including people with disabilities among theparticipants, and the burdens associated with participation, income-constrained householdswith disabled members seem likely to be underrepresented in the data. If households withdisabled members face greater barriers to accessing affordable foods or have less ability toengage in economizing behaviors, their under-representation in the Nielsen data wouldunderestimate the food costs faced by low-income families.

Households with incomes below the poverty line are much more likely to have lowereducation and literacy levels (see Figure 2 below), and more likely to speak a primarylanguage other than English. These households seems less likely to be represented in the

Nielsen dataset, and, as with households with disabled members, may face greater barriers toaccessing affordable foods or have less ability to engage in economizing behaviors.

As Jansen (2010) finds, lower-income households are less likely to use the internet, haveaccess at home to a broadband connection, and use email. F6F For example, 95 percent ofadults with incomes above $75,000 use the internet and 87 percent have broadband access athome; by comparison, only 57 percent of adults with incomes below $30,000 a year use theinternet, and only 40 percent have broadband access at home. Given the heavy reliance ofNielsen on email and web-based recruitment, internet users may be overrepresented amonglow-income households in the survey at a greater rate than among higher-incomehouseholds.

5 Similarly, data from the 2000 Census shows that the poverty rate for families that include a member with a disabilitywas substantially higher (12.8 percent) than it was for families without members with a disability (7.7 percent). Formore on the relationship between disability and poverty, see Fremstad (2009).

6 Jansen also finds they are less likely to pay bills, bank, and research and buy products online.


12/29


FIGURE 2

Prose Literacy Level by Income Group, 2003

229 237244

257268

282

303316

0

50

100

150

200

250

300

350


13/29


FIGURE 3

Percent of Households who Moved in 2008 and 2009, by Income Group

21.4%

17.0%

14.4%15.3%

14.5%15.1%

13.9% 14.2% 14.0%

12.5%

10.9%10.3%

9.7%

8.6%

7.0%

0%

5%

10%

15%

20%

25%

Unde

r$5

,000

orloss

$5,000

to$9

,999

$10,00

0to

$14

,999

$15,00

0to

$19

,999

$20,00

0to

$24

,999

$25,00

0to

$29

,999

$30,00

0to

$34

,999

$35,00

0to

$39

,999

$40,00

0to

$44

,999

$45,00

0to

$49

,999

$50,00

0to

$59

,999

$60,00

0to

$69

,999

$70,00

0to

$84

,999

$85,00

0to

$99

,999

$100

,000

and

ove

r

Household Income

Moved

between

2008

and

2009

Source: U.S. Census Bureau, Current Population Survey, 2009 Annual Social and Economic Supplement.

Thus, the methodology used by Broda et al. will likely select the subset of lower-income householdswho are careful comparison shoppers, and exclude those lower-income households who are likely tohave the greatest barriers to economizing on food, particularly those with disabilities, low literacylevels, and limited or no access to the internet. These differences alone could account entirely forBroda et als main finding that lower-income households pay less than higher-income ones for food.

Surprisingly, there is little to no research directly addressing the representativeness of the Nielsendata. As Zhen et al. (2009, p. 472) note, despite its potential for greater use, relatively little is knownabout the properties of household-based scanner data. In their research, Zhen et al. begin toaddress this research gap by comparing food expenditures in the Nielsen data with those in theBureau of Labors Consumer Expenditure Diary Survey. Looking first at the demographics of thetwo surveys, they note that the CES has more of the following groups than Nielsen: households

with children, Hispanic households, households with female heads not employed for pay, andhouseholds with female heads below age 35.

Using unpublished statistics provided by Zhen et al., Table 2 shows these differences for both weighted and unweighted samples, as well as differences in income groups between Nielsen andCES. As this table shows, households with incomes under $10,000 are considerablyunderrepresented in the Nielsen data compared to the CES. This is the case for both the unweightedand weighted versions of the datasets. The percentage of households with income below $10,000 in


14/29


the unweighted Nielsen data is less than half the percentage of such households in the CES; even inthe weighted data it is less than two-thirds of the CES. Similarly, higher-income householdsthegroup with incomes over $70,000are underrepresented in the Nielsen data compared to the CES.

TABLE 2

Differences between Nielsen and CES Data

Unweighted Data Weighted Data

HomescanConsumer

ExpenditureSurvey

Ratio ofHomescan

to CESHomescan

ConsumerExpenditure

Survey

Ratio ofHomescan

to CES

Household size 2.36 2.503 2.554 2.483

Presence of children 24% 33% 0.73 33% 32% 1.03Race

Hispanic 6% 12% 0.54 10% 11% 0.91White 83% 84% 0.99 79% 83% 0.95Black 10% 11% 0.92 12% 12% 0.96Asian 2% 4% 0.59 2% 3% 0.64

Presence of femalehead

90% 82% 1.1 81% 82% 0.99

Female head withcollege degree

32% 31% 1.04 20% 30% 0.67

Female head notemployed for pay

38% 47% 0.8 35% 48% 0.73

Female head below age35

7% 19% 0.35 17% 19% 0.89

IncomeBelow $5,000 0.8% 3.3% 0.24 1.3% 3.2% 0.41

$5,000$9,999 2.4% 5.2% 0.46 3.7% 5.5% 0.67$10,000$14,999 4.8% 6.7% 0.72 6.7% 7.0% 0.96$15,000$19,999 5.6% 6.2% 0.9 7.4% 6.5% 1.14

$20,000$29,999 14.8% 12.0% 1.23 17.5% 12.2% 1.43$30,000$39,999 14.8% 11.4% 1.3 15.0% 11.5% 1.3$40,000$49,999 13.3% 10.2% 1.3 12.4% 10.3% 1.2$50,000$69,999 19.5% 15.0% 1.3 16.7% 15.0% 1.11$70,000 and over 24.1% 30.0% 0.8 19.3% 28.8% 0.67

Finally, it is worth noting that the limited research that has examined the accuracy of the food-purchase data in the Nielsen data raises additional concerns about data quality.

Zhen et al. found significant underreporting of certain categories of food purchases inHomescan compared to the CES. The largest differences in percentage terms are for meats(beef, poultry, and pork), fresh fruits, eggs, fish and seafood, and fresh vegetables. Nielsenexpenditures are roughly 50 percent less for all of these categories than for the CES. In ananalysis of demographic factors that contribute to these differences, they found that certainhouseholds who are likely to purchase greater quantities of foodhigher-incomehouseholds and households with more memberstended to have larger expendituredifferences across the datasets.

Einav et al. (2008) examine the extent to which Nielsen participants accurately recordpurchases. Matching the purchase data recorded by households with sales data from a large


15/29


grocery retailer, they find recording errors in several dimensions. For some 20 percent offood-shopping trips recorded in the Nielsen data, they find no corresponding transaction inthe retailers data, suggesting that either the store or date information was recorded witherror. For trips that did match, roughly 20 percent of items purchased were not recorded.Price did not match for almost half the items that were recorded, with much of the

difference due to transactions that involved promotional or other temporary sale prices ineither the Nielsen or retailers data.

EvenIftheNielsenDatasetisRepresentativeofLow

IncomeShoppers,CanWeConcludethatthePoorPay

LessforFood?

Even if we assume that Nielsen households are representative of households in the population as a

whole, Broda et al.s broad interpretation of their results as showing that the poor pay less thanhigher-income households for food is overstated for at least two reasons.

First, in addition to the income-price differences summarized earlier, their results also show thathouseholds in the two lowest-income categories in the Nielsen data set (households with incomesunder $5,000 and those with incomes between $5,000 and $7,999) appear to pay more for food thanmost other low- and moderate-income households with incomes just above them. We cant saybased on the results they present whether these two income categories in the Nielsen datasetrepresent only a modest percentage of the officially poor or a substantial one. F7F But, if we assumerepresentativeness (and thats a big if) it appears that at least the poorest poor pay more for foodthan other low- and moderate-income households.

Second, Broda et al.s analysis doesnt account for additional costs that are likely to bedisproportionately borne by lower-income, economizing householdsincluding time spentcomparison shopping and time spent traveling to more distant stores or multiple storesor for theadditional benefits related to store amenities that are almost certainly disproportionately realized byhigher-income households. Evidence of the additional costs of paying less comes from Aguilarand Hurst (2005), who use American Time Use Survey data and a subset of Nielsen data to showhow households substitute time for money through shopping and home production. They find thathouseholds that shop (much) more intensively pay (somewhat) lower prices for identical food items:a doubling of shopping time lowers the price paid for a specific food item by about 10 percent.Thus, paying [a bit] less at the till entails paying more in other ways. The amount a household

7 One potential way to assess this question is to compare the income cutoffs for the bottom quintile in surveys such asthe CPS or CES with the cutoffs in the Nielsen data. In the 2005 Consumer Expenditure Survey, the bottom quintileof consumer units had pre-tax money incomes below $9,672 in 2005, which suggests that the below-$8,000households in the Nielsen data are akin to the bottom quintile. However, in the Current Population Survey, the bottom20 percent of households had pre-tax money incomes below $19,178, which suggests that the below $8,000 categoryin the Nielsen data accounts for a lesser share of poor households. However, given the considerable limitations of theNielsen income variable and the likelihood that the dataset isnt representative of low-income households, we cantreally draw any firm conclusions here.


16/29


pays at the grocery store till is lower, but the real cost of the item, a cost that adds in search costs,including time and travel, may not be much different and may even be higher.

As for the additional benefits related to store amenities, Broda et al. acknowledge that store qualitydifferences explain at least a portion of the difference in price, but their estimate that these

differences account for only about a third of the differences in the prices of identical items likelylowballs the effect of store quality. They attempt to take store quality into account by controlling forretail chain, but it seems quite likely that there are considerable quality differences between storeswithin retail chains, and that these differences are associated to some extent with income differences.For example, the quality of grocery stores in the same retail chain may be higher in higher-incomeneighborhoods than in lower-income ones.

In short, even assuming that Nielsen households are representative of households in the population,we should be very cautious about concluding that the income differences Broda et al. find in theprices paid per food item correspond to income differences of a similar magnitude in the welfare orliving standards associated with those items.

A related concern is that Nielsen participants may be more likely to record purchases from storesthat participate in Nielsens Scantrack systembecause item pricing data for these stores is suppliedby Nielsen and doesnt need to be provided by the consumerand less likely to record purchasesfrom (generally more expensive) smaller stores. As a result, Broda et al. may underestimate theimpact of higher prices in these stores on the overall average food prices paid by low-incomeconsumers.

IncomeDifferencesinFoodPricesandInequality

Broda et al. conclude that if the poor can access the same goods as the rich at a lower price, the useof common price indexes between income groups misses an important difference in [their]relative real incomes.... Using varying assumptions about the extent to which the lower prices paidby moderate-income households reflect an income gain, they suggest income inequality may bebetween 2.5 to 5 percent less than shown by national statistics.

However, stating that the poor can access the same goods as the rich at a lower price implies thatthere is some binding constraint on the ability of the rich to shop at the same place, and in the same way, as most other households dothat is, that the rich cannot access the same goods as otherhouseholds at the same price. This, of course, is absurd. Given their greater economic resources, therich have considerably more capacity than households with lower incomes to engage in economizingshopping behaviors. That they more likely prefer to spend their leisure time and greaterconsumption power in other ways than economizing on food does not imply that they are really notas well-off compared to lower-income families as normal income data shows.

Similarly, instead of viewing income differences in food prices as a factor that moderates the reallevel of economic inequality, it seems more reasonable to view the massive increase in income andwealth inequality over the last several decades as factors that have directly contributed to the incomedifferences in food prices. As Figure 4 below shows, between 1979 and 2006, average after-taxincome of households in the top quintile of the income distribution (households earning roughly


17/29


$70,000 or more in 2006) increased by 86 percent, an increase almost eight times that of the bottomincome quintile, four times that of the second and middle quintiles, and about two and a half timesthat of the fourth quintile. Comparing this chart with Figure 1the chart reproduced from Broda etal.showing the relationship they find between income and prices is quite instructive. The high-income groups that are paying more for food in the Nielsen dataset (those with incomes between

$70,000-$99,999 and with incomes of $100,000 and more) are roughly the same groups that sawtheir incomes increase at more than four times the rate of typical households.

FIGURE 4

Increase in After-Tax Income by Income Group, 1979-2006

1118 21

32

86

112

143

256

-

50

100

150

200

250

300

LowestQuintile

SecondQuintile

MiddleQuintile

FourthQuintile

HighestQuintile

Top 10% Top 5% Top 1%

Percent

Source: Congressional Budget Office (2009).

As a result, low- and moderate-income households have likely felt greater pressure to economize infood and other areas than they would have if their incomes had grown at a similar rate as they didfor higher-income households. In addition, it seems quite likely that the increase in incomeinequality has resulted in a corresponding increase in what might be called quality inequality. Inother words, there have been greater store and item quality improvements for higher-incomeconsumers who have the experienced the greatest gains in income than for lower-income shoppers

who have seen quite modest ones.

Also relevant here are the findings of Frankel and Gould (2001), which looks at general retailprice differences (groceries, transportation, health, and miscellaneous goods) between cities andhow they relate to income. They find that an increase in the presence of lower-middle incomehouseholds, relative to poor or upper income households, is associated with lower prices. Forexample, they find that:


18/29


... supermarkets charge the lowest prices in communities where a greater proportionof households have an income level between one and two times the poverty line. Ifone percent of these lower to middle-income households fall into poverty ... foodprices rise by .85 percent. Likewise, if that same one percent of households increasestheir income to above twice the poverty rate, food prices rise by 0.52 percent.

Frankel and Gould conclude that higher prices are a result not of greater poverty per se, butrather of an absence of lower-middle income households. Thus, the poor pay more ... butonly to the extent they are isolated from lower-middle income households.

F

8FThe findings of

Broda et al. are not necessarily inconsistent with those of Frankel and Gould. Broda et al. findthat lower-middle income households pay less for food than very-low income ones, and that African-American households pay slightly more (.2 percent) than white households. Thesehigher prices for very-low-income households and African Americans could be driven by higherlevels of geographic segregation for these groups.

TheQuestion

of

Whether

the

Poor

Pay

More

Today:

Food,TransportationandFinancialServices

Broda et al. generally summarize prior research as finding that the poor indeed pay more thanhouseholds of higher income for the goods and services they purchase. This general conclusion isreferred to throughout the article as being the conventional wisdom. Yet, the research on foodprices is far more nuanced than Broda et al. imply. More generally, at least two aspects of the overallquestion of whether the poor pay more have implications for the cost of food understood broadly(including the cost of shopping for it): 1) the cost of transportation, and the increasing share offamily budgets going to transportation; and 2) costs related to the use of credit cards and otherfinancial services.

Food

On the specific subject of whether the poor pay more for food, the research has been more variedthan Broda et al. imply, a fact that has been widely noted by other researchers in the field. Forexample, Hayes (2001) explains that research is decidedly mixed on the question of whether pricesare higher in poor, urban neighborhoods, and Frankel and Gould (2001) conclude that priorresearch found only a small or insignificant relation between poverty and prices.

Similarly, in 1997, analysts at the U.S. Department of Agricultures Economic Research Service(Kaufmann et al.) summarized the research (at that point, some fourteen studies that met their

criteria for relevance and soundness of methods) and conducted their own additional research usingfood stamp redemption and other data. On the one hand, they found that low-income householdsface slightly higher food prices than other households for the same food items, with the likelyaverage difference being less than one percent, although the difference could average up to 3.1percent higher ... in some low-income neighborhoods .... On the other hand, based on analyses ofsurveys of household food consumption and expenditures, they concluded that low-income

8 For a paper that provides a theoretical model for these results, see Somekh (2010).


19/29


households actually spend less on a per unit basis for the foods they buy, largely because they selectmore economical foods such as store label and generic items, larger package sizes, and lower qualityitems in order to realize lower food costs. In essence, if low-income households bought the sameitems and shopped in the same way as other households, they would pay more. But because theybuy different items and engage in other economizing behaviors, they spend less for food, on

average, compared with all households.

Subsequent research fills out this picture, but does not change it significantly. For example, Stewartand Blisard (2007), who also use Nielsen data, examine how expenditures on vegetables (both totaland per serving) differ by household income and community characteristics. Consistent withKaufmann et al. (1997) they find that the lower a households income, the more it tries to hold downunit vegetable costs with economizing practices, such as buying vegetables that cost less per servingor more frequently buying items that are on sale. However, consistent with Frankel and Gould(2001), they also find that households living in communities with high real estate prices, or with ahigh incidence of poverty tend to pay more.

Much recent research focuses less exclusively on whether the poor pay more for food, and more

broadly on whether low-income and other households have sufficient access to food that is bothhealthy and affordable. For example, in a comprehensive review of more than 100 studies on foodaccess and pricing, Treuhaft and Karpyn (2010) focus on the extent of inequitable access to healthyfoods, and the relationship of access to healthier eating and better health overall. And, a 2009 USDAreport finds that 23.5 million peoplejust over eight percent of the U.S. populationlive in low-income areas (areas where more than 40 percent of the population has income at or below 200percent of the poverty line) that are more than one mile from a supermarket. People living in theseareas spend about 25 percent more time (per trip) traveling to a grocery store than the nationalaverage.

Finally, whether or not the poor pay more for food purchased for consumption at home is today a

relatively small part of the broader and now decades-old question of whether the poor pay more.In 1960, food expenditures accounted for about one-quarter of the average familys expenditures;today, they have dropped to nearly half thatabout 13 percent of the average familys expenditures.The food expenditure share of low-income households today (16.2 percent) is slightly higher thanthe average for all households, but the trend is the same.

Transportation

Research cited earlier suggests that households can economize on food if they shop more intensivelyand travel farther to find bargains. Unlike food, transportation has grown as a share of familybudgets in recent decades, going from 14.7 percent of the average familys budget in 1960 to 19percent in 2002-2003 (BLS, 2006). Today, the average low-income family spends more of its budget

on transportation than it does on food at home.

Research on transportation price differences by income is limited, but suggests that low-incomepeople face higher costs to own and operate a car. Myers et al. (2009) find that gasoline prices arehigher in poor neighborhoodsa ten percentage point increase in the percent of families withincomes below the poverty line relative to families with incomes between one and two times thepoverty line, is associated with a 0.7 percent increase in retail gas prices. Morton et al. (2001) findthat neighborhood income, race and other factors associated with household income (including


20/29


educational attainment and renter status) are related to car purchase prices. Based on Mortonsresearch, Fellowes (2006, p. 35) explains:

... two customers who buy the exact same car will pay different prices that varysystematically with certain characteristics. We can see the power of these effects by

comparing two hypothetical car buyers. The first is white, has a high school diploma,owns a house, and lives in a neighborhood with a median income of $80,000. Thesecond customer is black, dropped out of high school, rents, and lives in aneighborhood with a median income of income of $20,000. According to theanalysis of Morton and her colleagues, the second customer will pay about $500more than the first customer for the exact same car.

Low-income car owners also frequently pay more to finance a car purchase (see Fellowes (2006) andVan Alst (2009)) and for car insurance (see, e.g., State of Missouri (2004), Fellowes (2006), and Wu(2007)).F9

FinancialServices

The cost of credit and other financial services has been a common focus of research since Caplovitz(1963) documented the higher prices paid for retail installment credit by low-income families in New York City. Research conducted since then generally finds that lower-income families continue tohave less access to credit and pay more for it. F10F What has changed over the last several decades is thegrowth of financial services as a sector of the economy and increased importance of financialproducts in everyday expenditures. F11F Two aspects are particularly relevant to food cost.

First, researchers have consistently found that low-income households generally have less access tomainstream financial services, and are more likely to be unbanked than higher-incomehouseholds. F12F According to the FDIC (2009), about 20 percent of households with incomes below

$30,000 lack a checking or savings account. Many of these unbanked households pay high fees (tocheck cashers, payday lenders, and other non-mainstream financial service providers) simply toaccess their income. As a result, as Barr (2004) concludes, they face high costs, relative to theirincome, for basic financial services.

9 Home and auto insurers increasingly use credit scores to determine premium rates, a practice that results in higherrates for lower-income families on average, even through there is causal explanation for a linkage between creditscores and insurance risk.

10 For more mainstream forms of credit, income and credit scores (and credit history more generally) determine bothaccess to various credit products and the price (the interest rate and other charges) that households pay for them.For example, for persons seeking a 30-year mortgage for a $175,000 home, annual mortgage costs increase as creditscores decline. Based on current rates, households with a credit score of 650 will pay about $1,700 more a year forsuch a mortgage than those with scores above 750. Simlarly, Bostic et al. (2004) find that credit scores in 2001 were

positively correlated with income. In regression analyses that control for a range of demographic factors, they findthat individuals in the bottom two income quintiles have credit scores significantly lower than the scores of those inthe top two income quintiles, and are also much more likely to have scores below 660. A related concern is thatsome contract features in the credit-card and subprime mortgage markets may induce consumers to borrow toomuch and to make suboptimal contract and repayment choices. Heidhues and Koszegi (2010). Low- and middle-income consumers may be particularly disadvantaged by these features.

11 On the former, see, e.g., Johnson (2009); on the latter, see Evans and Schmalensee (2005).12 FDIC (2009) finds that 27.1 percent of households with incomes under $15,000 a year are unbanked, and the

likelihood of being unbanked declines rapidly as income increase; the vast majority of unbanked households (71percent) have incomes below $30,000. p. 11.


21/29


Second, higher-income households likely benefit disproportionately from the now-widespreadacceptance of credit cards by stores selling groceries. The costs to food retailers associated withcredit card acceptance (about $2 for every $100 of purchases) are passed on in the form of higherprices for everyone (grocery stores dont provide a discount for shoppers who pay cash). F13F

To the extent that higher-income households are more likely to use credit cards for grocerypurchases than lower-income householdswhich seems almost certain given general data onincome differences in the use of credit cards (see Mann 2009)generally poorer customers whodont use credit cards to purchase groceries are subsidizing the purchases of generally richercustomers who do. F14F In addition, as GAO (2009) finds, the fees that credit card companies like Visaand Mastercard charge retailers are generally higher for their premium cards (ones that providecash-back, frequent flyer miles, and other bonuses) than for their basic cards. Lower-incomeshoppers are probably less likely to hold these cards and receive the benefits associated with them,but still subsidize their use by higher-income households.

Arethe

Current

Federal

Poverty

Thresholds

(Way)

Too

High?

In a short section of their paper, Broda et al. claim that the actual poverty rate was about fourpercent in 2006 rather than the official rate of just over ten percent, and that the poverty rate fell byabout 60 percent between 1970 and 2006 rather than remaining stable. This issue is distinct from thequestion of whether the poor pay more for food, and draws on analyses by Broda and Weinstein(2008) and Broda and Weinstein (2010), as well as various conceptual and technical assumptions thatare highly contested among inflation experts. As such, it is only discussed briefly here.

The current poverty measure was developed in the early 1960s, and has been adjusted since 1969using the CPI. Broda et al. arrive at a four-percent poverty rate by replacing the CPI with their ownprice index, which adjusts for new goods, quality, and substitution biases they argue are present inthe CPI. Whatever the technical merits of the large quality and substitution adjustments to the CPIproposed by Broda and Weinstein (they are dubious, but a discussion of them is beyond the scopeof this article)F15F, applying them retroactively to nearly four decades of income data produces resultsthat are simply not credible. Broda et al.s estimated poverty rate of four percent in 2006 is roughlyequal to the 4.2 percent of people in families that year who lived below half of the official poverty

13 On this point, see U.S. Government Accountability Office (2009), which notes that increased merchant costs forcard transactions may lead to higher prices for non-cardholding consumers.

14 Credit card use is no longer rare among lower-income households. According to Mann (2009) about 30 percent of

households in the bottom quintile held credit card balances in the 2004 Survey of Consumer Finances, compared toabout 55 percent of those in the middle quintile. But this usage generally comes at a greater cost for lower-incomehouseholds than for higher-income ones. Wheary and Draut (2007) find that the percentage of credit card holderswith high-interest rates (more than 20 percent APR) declines as household income rises. Only 6.7 percent ofcardholders with household incomes between $50,000 and $100,000 pay very high interest rates, compared to 15percent of those with incomes below $10,000 and 12 percent of those with incomes between $10,000 and $25,000.

15 For example, Broda et al. would reduce the poverty threshold to account for what they argue is a reduction in theprice level caused by the introduction of new goods. But a 2002 National Academy of Sciences (2002) panel on theCPI concluded that virtual price reductions associated with the introduction of new goods should not be imputed foruse in the CPI. For background on debates over the accuracy of the CPI, see Baker (1998).


22/29


thresholds. Thus, Broda et al.s adjusted poverty thresholds would be equal to half the currentofficial thresholds. In 2006, the official poverty line for a family of two adults and two children was$20,444according to Broda et al., it should have been set at $10, 222.

When the poverty line was initially developed in the 1960s the thresholds were set equal to the value

of USDAs Thrifty Food Plan (then know as the Economy Food Plan) multiplied by three (based ona finding that the average family spent about one-third of its income on food). USDAs Thrifty FoodPlan represents a minimal-cost, nutritious diet, one that assumes (unrealistically) that all meals afamily eats are prepared at home. According to USDA, the annual cost of the thrifty plan for afamily of four in 2006 was $5,341 ($445 a month or $3.70 per person per day) if the children areboth under age six and $6,217 ($518 a month or $4.31 per person per day) if the children are bothover age six. Subtracting the latter food-budget amount from Broda et al.s adjusted 2006 povertythresholds (roughly $852 a month) would mean a two-adult, two-child (both over age six) family with income equal to the Broda poverty line would have been left with $4,005 in 2006$334 amonthto pay for safe and decent housing, transportation, clothing, health care, child care andother work expenses, and all other necessities. Outside of rare circumstances, this would have been apractical impossibility for a couple raising two children.F16F

The real problem with the current poverty line is that it has not been adjusted over time to reflectchanges in minimally adequate living standards, and, as a consequence is too low. This problem is theresult of the decision made by the Nixon Administration in 1969 to only adjust the poverty line forprice inflation, instead of updating it in a way that better reflected changes in mainstream livingstandards. Mollie Orshansky, who originally developed the measure, criticized this decision becauseit would freeze the poverty line despite changes in buying habits and changes in acceptable livingstandards (Fisher 1997).

Figure 5 below shows how the Nixon Administrations decision has had the effect of definingincome deprivation down over the past half century. In 1963, the poverty line for a family of four

was $3,128. This was equal to about 44 percent of median income for a family of four that year, andabout 72 percent of the amount the average amount that Americans told Gallup a family of fourneeded to get along at a minimum level. By 2007, the federal poverty line for a family of four($21,203), adjusted only by the CPI, had fallen to 28 percent of median income for a family of fourthat year and about 41 percent of the average minimum get along level in the 2007 Gallup survey.

The adjusted poverty thresholds produced by Broda et al. would fall even further below mainstreamliving standards today. Their adjusted poverty threshold for a family of four would amount to 14percent of median income in 2007 and 20 percent of the average amount that Americans toldGallup was needed in 2007 to get along at a minimum level.

16 The bottom line would be roughly similar in 2010. The Broda et al. poverty line would be about $11,025 for a familyof four in 2010 (half of the HHS poverty guideline of $22,050). The thrifty food plan cost this year for a family offour is $6.084 (both children under age 6) or $6,984 (both children over age 6), leaving the latter family with $337 amonth to pay for all non-food necessities.


23/29


FIGURE 5

Change in the Federal Poverty Live, Minimum Get-Along Average, and Median Income, 1963-2007

1963

$3,128

$4,328

$7,138

0

1000

2000

3000

4000

5000

6000

7000

8000

Federal Poverty Line Gallup Minimum Get-

Along Average

Median Income

1963

dollars

FPL equal to72% of

Gallup "get-

along"

minimum

FPL equalto 44% of

median

income

2007

$21,203

$52,087

$75,675

0

10000

20000

30000

40000

50000

60000

70000

80000

Federal Poverty Line Gallup Minimum Get-Along Average

Median Income

2007

dollars

FPL equal to41% of

Gallup "get-

along"

minimum

FPL equal to28% of

median

income

While Broda et al.s analysis implies that the price level effectively only moves downward in responseto the introduction of new goods, a more balanced view would recognize that the introduction and widespread diffusion of new goods increases the cost of maintaining a standard of living that isminimally adequate in todays terms. As Gordon Fisher (2006) explains:


24/29


As technology progresses and the general standard of living rises, new consumptionitems are introduced. They may at first be purchased and used only by upper-incomefamilies; however, they gradually diffuse to middle- and lower-income levels. Thingsoriginally viewed as luxuriesfor instance, indoor plumbing, telephones, and

automobilescome to be seen as conveniences and then as necessities. In addition,changes in the ways in which society is organized (sometimes in response to newnecessities) may make it more expensive for the poor to accomplish a given goalas when widespread car ownership and increasing suburbanization lead to adeterioration in public transportation, and the poor are forced to buy cars or hiretaxis in order to get to places where public transit used to take them. Finally, thegeneral upgrading of social standards can make things more expensive for thepooras when housing code requirements that all houses have indoor plumbingadded to the cost of housing.

A related point is that because inflation-adjusted poverty measures are relative to the specific pastbase years (and social and economic contexts) in which they were established, their value today will

vary depending on the base years they were initially set in. Suppose, for example, that the unofficialpoverty line used by the Subcommittee on Low-Income Families in 1949, which was equal inconstant dollars to about 85 percent of the official poverty line in 1964 (Fisher 1995), had beenadopted at that time as the nations official poverty measure and adjusted for inflation in subsequentyears. In 2010, the threshold for this base-1949 poverty line would be considerably lower than thethresholds for the actual official (base-1963) line. Excepting perhaps Broda et al., most wouldagree that the base-1963 line better reflects a current poverty-level income than the base-1949 line,but this then begs the question of why the measure for the current poverty level should be based onany previous year several decades earlier (and determinations of minimum income adequacy madeby officials who are now long gone) rather than relatively current data and determinations.

Poverty measures that are set based on some minimally acceptable level of income in a base year andthen only adjusted in subsequent years by a price index may have some utility if the goal is to tracktrends in income deprivation over some fairly short time frame (say five-ten years) when there islittle change in the economic and social context. But they become increasingly uninformative andunreflective of prevailing real-world economic conditions when used for longer time frames,especially during periods in which there has been considerable economic progress and innovation.

There are at least two solutions to this problem, both of which involve moving away from aninflation-adjusted measure as the primary measure of poverty. The first was proposed by a panel ofthe National Academy of Sciences in the mid-1990s, and proposed for adoption (in a modified version known as the Supplemental Poverty Measure or SPM) earlier this year by the Obama

Administration. The SPM would initially set the poverty line equal to expenditures on food, clothing,shelter and utilities at the 33rd percentile multiplied by 1.2. Instead of an inflation adjustment, thisintial threshold would be recalculated each year by applying this same formula to the most recentexpenditure data. The second solution, adopted earlier this year by the United Kingdom, would setthe poverty line equal to percentage of median (60 percent) income, and then adjust over time toreflect that percentage. The UK approach also incorporates two additional measures of poverty: 1)one that is set at 60 percent of median income in a base year and then adjusted for a ten-year periodby inflation; and 2) a material deprivation index composed of various direct measures of materialhardship that reflect prevailing community norms about the necessities needed for a minimally


25/29


decent standard of living. These two solutions are not inconsistent with one another. The bestpoverty measurement framework for the U.S. would, like the UK approach, include three or fourcore measures of economic deprivation. F17F A version of the SPM could be substituted in place of aninflation-adjusted measure in this kind of broader framework.F18F

Conclusion

Given the likelihood that low-income households are not adequately represented in the Nielsendataset, it should not be used to draw conclusions about the relative prices paid by these households(or other economically disadvantaged groups). USDA and any other federal agencies that useNielsen data should carefully examine its representativeness before using it in any future research.

Even if we assume that the Nielsen data are representative, what it actually shows is morecomplicated than the rather simplistic interpretation offered by Broda et al. that the poor pay less.In fact, their data show that a substantial percentage of low-income households in the dataset pay

somewhat more for food than moderate-income households, and that the Gilded-Age level ofinequality that prevailed during the last decade is reflected in shopping for the most basic ofcommodities.

17 Canada also uses three distinct measures of income deprivation: 1) a low-income cutoff (LICO) that represents anincome threshold below which a family is likely to spend significantly more of its income on food, shelter andclothing than the average family; 2) a low-income measure (LIM) defined as 50 percent of median adjusted family

income; and 3) a market basket measure (MBM) that represents the cost of a basket of goods and services that wouldallow a family to eat a nutritious diet, buy clothing for work and social occasions, house themselves in theircommunity, satisfy basic transportation needs for work, school, shopping and participation in community activitiesand pay for other necessary expenses. See Statistics Canada (2010) for the most recent estimates using thesemeasures.

18 A limitation of the SPM is that, as currently proposed, it will be set at a level that bears little resemblance to theamount needed to obtain a minimally adequate standard of living. Moreover, it will only reflect changes over time infood, clothing, shelter, and utility consumption, and not the broader (and generally expanding) market basket ofgoods and services that are needed for a minimally decent standard of living, including transportation, health care,and savings for basic retirement security. For more on the SPM and poverty measurement, see Fremstad (2010).


26/29


References

Aguiar, Mark, and Erik Hurst. 2005. Lifecycle Prices and Production. Federal Reserve Bank ofBoston, Public Policy Discussion Paper 05-3.http://www.bos.frb.org/economic/ppdp/2005/ppdp053.pdf

Baker, Dean (ed). 1998. Getting Prices Right: The Debate Over the Consumer Price Index. Economic PolicyInstitute. M.E. Sharpe.

Barr, Michael S. 2004. Banking the Poor. Yale Journal on Regulation21(1): 121-237.

Bostic, Raphael W., Paul S. Calem, and Susan M. Wachter. 2004. Hitting the Wall: Credit as anImpediment to Homeownership. BABC 04-5, Joint Center for Housing Studies, HarvardUniversity. http://www.jchs.harvard.edu/publications/finance/babc/babc_04-5.pdf

Broda, Christian, Ephriam Leibtag, and David E. Weinstein. 2009. The Role of Prices in Measuring

the Poors Living Standards.Journal of Economic Perspectives, 23(2): 77-97.http://pubs.aeaweb.org/doi/pdfplus/10.1257/jep.23.2.77

Broda, Christian and David E. Weinstein. 2008. Prices, Poverty, and Inequality: Why Americans Are BetterOff than You Think. Washington, DC: AEI Press.

Bureau of Labor Statistics. 2006. 100 Years of U.S. Consumer Spending: Data for the Nation, NewYork City, and Boston. BLS Report 991. http://www.bls.gov/opub/uscs/report991.pdf

Caplovitz, David. 1963. The Poor Pay More: Consumer Practices of Low-Income Families. New York: FreePress.

Chung, Chanjin, and Samuel L. Myers, Jr. 1999. Do the Poor Pay More for Food? AnAnalysis of Grocery Store Availability and Food Price Disparities.Journal of Consumer Affairs,33(2): 27696.

Congressional Budget Office. 2010. Using a Different Measure of Inflation for Indexing FederalPrograms and the Tax Code. Washington, DC: CBO.http://www.cbo.gov/ftpdocs/112xx/doc11256/CPI_brief.pdf

Congressional Budget Office. 2009. Data on the Distribution of Federal Taxes and HouseholdIncome. April. http://cbo.gov/publications/collections/tax/2009.cfm

Einav, Liran, Ephraim Leibtag, and Aviv Nevo. 2008. On the Accuracy of Nielsen HomescanData. ERR-69, U.S. Dept. of Agriculture, Economic Research Service.

Evans, David S. and Richard Schmalensee. 2005. Paying with Plastic: The Digital Revolution in Buying andBorrowing. Cambridge, MA: MIT Press.

Federal Deposit Insurance Corporation. 2009. National Survey of Unbanked and UnderbankedHouseholds. Washington, DC: FDIC. http://www.fdic.gov/householdsurvey/full_report.pdf


27/29


Fellowes, Matt. 2006. From Poverty, Opportunity: Putting the Market to Work for Lower IncomeFamilies. Washington, DC: The Brookings Institution, Metropolitan Policy Program.http://www.brookings.edu/~/media/Files/rc/reports/2006/07poverty_fellowes/20060718_PovOp.pdf

Fisher, Gordon. 1997. Poverty Lines and Measures of Income Inadequacy in the United States

Since 1870: Collecting and Using a Little-Know Body of Historical Material. Paper presented to the22nd Meeting of the Social Science History Association, October 17, 1997 in Washington, D.C.http://aspe.hhs.gov/poverty/papers/since1870.htm.

Fisher, Gordon M. 1995. Is There Such a Thing as an Absolute Poverty Line Over Time? Evidencefrom the United States, Britain, Canada, and Australia on the Income Elasticity of the Poverty Line.U.S. Department of Commerce, Census Bureau.http://www.census.gov/hhes/www/povmeas/papers/elastap4.html

Frankel, David M. and Eric D. Gould. 2001. The Retail Price of Inequality.Journal of UrbanEconomics, 49: 219239.

Fremstad, Shawn. 2010. A Modern Framework for Measuring Poverty and Basic EconomicSecurity. Center for Economic and Policy Research.http://www.cepr.net/documents/publications/poverty-2010-04.pdf

Fremstad, Shawn. 2009. Half in Ten: Why Taking Disability into Account is Essential to ReducingIncome Poverty and Expanding Economic Inclusion. Washington, DC: Center for Economic andPolicy Research. http://www.cepr.net/documents/publications/poverty-disability-2009-09.pdf

Harris, J. Michael. 2005. Using Nielsen Homescan Data and Complex Survey Design Techniquesto Analyze Convenience Food Expenditures. Economic Research Service, United StatesDepartment of Agriculture.

Hayes, LaShawn Richburg. 2000. Do the Poor Pay More? An Empirical Investigation of PriceDispersion in Food Retailing. Working Paper, Princeton University Industrial Relations Section.

Heidhues, Paul and Botond Koszegi. 2010. Exploiting Naivete about Self-Control in the CreditMarket.American Economic Review, 100: 2279-2303.

Jansen, Jim. 2010. Use of the Internet in Higher-Income Households. Pew Internet and AmericanLife Project.

Johnson, Simon. 2009. The Quiet Coup. The Atlantic. May.

http://www.theatlantic.com/magazine/archive/2009/05/the-quiet-coup/7364/

Kaufman, P., MacDonald, J., Lutz, S. and Smallwood, D. 1997. Do the Poor Pay More for Food?Item Selection and Price Differences Affect Low-Income Household Food Costs. AgriculturalEconomics Report No. 759. Washington, DC: Economic Research Service, United StatesDepartment of Agriculture.


28/29


Mann, Ronald J. 2009. Patterns of Credit Card Use Among Low- and Moderate-IncomeHouseholds. In Insufficient Funds: Savings, Assets, Credit and Banking Among Low-Income Householdsedited by Rebecca M. Blank and Michael S. Barr. New York: Russell Sage Foundation.

Meyer, Bruce, Wallace Mok, and James Sullivan. 2009. The Under-Reporting of Transfers in

Household Surveys: Comparisons to Administrative Aggregates. NBER Working Paper 15181.Cambridge, MA: National Bureau of Economic Research.http://aida.econ.yale.edu/seminars/labor/lap10/Meyer-100423_1.pdf.

Morton, Fiona Scott. 2006. Consumer Benefit from Use of the Internet. In Innovation Policy and theEconomy, Volume 6edited by Adam B. Jaffe, Josh Lerner and Scott Stern. Cambridge, MA: MITPress.

Morton, Fiona Scott, Florian Zettelmeyer, and Jorge Silva-Risso. 2001. Consumer Information andPrice Discrimination: Does the Internet Affect the Pricing of New Cars to Women and Minorities?Working Paper 8688, National Bureau of Economic Research.

Myers, Caitlin Knowles, Grace Close, Laurice Fox, John William Meyer, and Madeline Niemi. 2009.Retail Redlining: Are Gasoline Prices Higher in Poor and Minority Neighborhoods? MiddleburyCollege, Economic Discussion Paper No. 09-06.http://sandcat.middlebury.edu/econ/repec/mdl/ancoec/0906.pdf

National Academy of Sciences. 2002.At What Price? Conceptualizing and Measuring Cost-of-Living andPrice Indexes. Washington, DC: National Academy Press.

She, Peiuyun and Gina A. Livermore. 2009. Long-Term Poverty and Disability Among Working-Age Adults.Journal of Disability Policy Studies, 19(4): 244-256.

Somekh, Babak. 2010. The Effect of Income Inequality on Price Dispersion. Working Paper,University of Oxford.

State of Missouri, Department of Insurance. 2004. Insurance-Based Credit Scores: Impact onMinority and Low Income Populations in Missouri.

Statistics Canada. 2010. Income in Canada, 2008.http://www.statcan.gc.ca/pub/75-202-x/75-202-x2008000-eng.htm.

Stewart, Hayden, and Noel Blisard. 2008. Who Pays More for Food?Journal of AgriculturalEconomics, 59(1): 150-168.

Treuhaft, Sarah and Allison Karpyn. 2010. The Grocery Gap: Who Has Access to Healthy Foodand Why It Matters. PolicyLink and The Food Trust.http://www.policylink.org/atf/cf/%7B97C6D565-BB43-406D-A6D5-ECA3BBF35AF0%7D/FINALGroceryGap.pdf

United States Department of Agriculture. 2009. Access to Affordable and Nutritious Food:Measuring and Understanding Food Deserts and their Consequences. Report to Congress.


29/29


United States Department of Education. 2007. Literacy in Everyday Life: Results from the 2003National Assessment of Adult Literacy. NCES 2007-480.http://nces.ed.gov/pubs2007/2007480.pdf

United States Government Accountability Office. 2009. Credit Cards: Rising Interchange Fee

Increased Costs for Merchants, but Options for Reducing Fees Pose Challenges. GAO-10-45.

Van Alst, John. 2009. Fueling Fair Practices: A Road Map to Improved Public Policy for Used CarSales and Financing. Washington, DC: National Consumer Law Center.Wheary, Jennifer and Tamara Draut. 2007. Who Pays? The Winners and Losers of Credit CardDeregulation. Washington, DC: Dmos.

Wu, Chi Chi and Birny Birnbaum. 2007. Credit Scoring and Insurance: Costing Consumers Billionsand Perpetuating the Economic Racial Divide. National Consumer Law Center and Center forEconomic Justice.http://www.cej-online.org/NCLC_CEJ_Insurance_Scoring_Racial_Divide_0706.pdf

Xiao, Jing Jian. 2006. Who are Potential Victims of Universal Default? Evidence from the 2004Survey of Consumer Finances.

Xiao, Jing Jian and Radovan Vadovic. 2005. The Cost of Deregulating the Credit Card Industry andits Implications for Consumers.

Zhen, Chen, Justin L. Taylor, Mark K. Muth, and Ephraim Leibtag. 2009. UnderstandingDifferences in Self-Reported Expenditures between Household Scanner Data and Diary SurveyData: A Comparison of Homescan and Consumer Expenditure Survey.Applied Economic Perspectivesand Policy[formerly Review of Agricultural Economics], 31(3): 470-492.

Income, Inequality and Food Prices

Documents

Transcript of Income, Inequality and Food Prices